Why it is needed
CAD systems describe parts with mathematical surfaces such as NURBS that represent any shape exactly [3]. Real-time graphics, by contrast, works with polygon meshes. The Library of Congress explicitly calls the breaking of a surface into small triangles or other polygons "tessellation" [2].
How it works
The Open CASCADE geometry kernel describes a typical two-step approach: first the edges of each face are discretised into points, then the face itself is triangulated [1]. Because neighbouring faces share the same edge points, no gaps appear.
The tolerance decides
Two values control the result [1]:
- Linear deflection: the maximum distance between the mesh and the true surface.
- Angular deflection: the maximum angle between consecutive segments.
Too coarse and round parts turn angular. Too fine and a part has millions of triangles. Good values depend on part size: a curved rail needs a different resolution from a hall frame.
A practical approach
Tessellate coarsely first, then zoom in on the most critical radii in a viewer and refine the tolerance only where facets become visible. Many tools allow different values per part. The model stays light without looking angular where people look.
Typical problems afterwards
- Many triangles on large flat areas, which decimation can remove again.
- Normals that are flipped or smoothed across hard edges.
- Hidden internal parts that were tessellated too and only add weight.
The biggest gain usually comes before tessellation: do not convert parts nobody will see. The guide CAD to web 3D covers the full route from STEP to GLB.